Observers and Diagnosis of Multi-Layer

 Condition Systems

 

Professor Larry Holloway

Department of Electrical and Computer Engineering

and

Center for Robotics and Manufacturing Systems

 

University of Kentucky

 

 

A condition system is a collection of discrete-event models that interact with each other and the external environment through condition signals. 

The model framework allows the representation of a plant and a controller

as a set of interacting component models with explicit cause and effect relationships.  In this talk, we discuss the fault diagnosis problem for such systems.  A fault is defined as an inconsistency between observed behavior versus expected behavior.  We show that each component model

represents a constraint on the condition-language of the system.  Fault diagnosis is then the identification of the component models for which relaxation of their language constraints account would resolve the discrepancy between observed and expected behavior.  To identify these

suspect component models, we convert the structure of the system into a

causal net.  Analysis of this causal net allows us to identify a superset of components whose behavior relaxation would account for the observed faulty behavior of the system.

 

 
Friday, March 8, 2002

3:30 - 5:00 p.m.

1500 EECS